Augmented reality/United Kingdom; Denmark
The AR trainees were faster on the model they trained on, and no better anywhere else
Thirty novice bronchoscopists were randomised to VR or AR training. Neither clearly won, and the one speed advantage was on the model that group had practised on.
Thirty novice bronchoscopists were randomised to virtual reality or augmented reality simulator training. VR was marginally better on diagnostic completeness on the first test model; AR was faster on the second model, which happened to be the one the AR group had trained on. VR training carried 15 NASA-TLX points less mental demand and 35 points more physical demand.
What happened
Flexible bronchoscopy means steering a camera through a branching airway and reaching all eighteen segments without getting lost. It is a skill that simulators teach well, and there are two kinds. Virtual reality simulators put the whole airway on a screen. Augmented reality simulators overlay the anatomy onto a physical model the trainee is actually holding.
Which is better was unknown. The authors of this trial – from Imperial College London’s APMIC group in Surgery and Cancer, the Copenhagen Academy of Medical Education and Simulation at the University of Copenhagen, and Chelsea and Westminster Hospital NHS Foundation Trust – say so directly.
They randomised thirty novice bronchoscopists, fifteen to each, and tested them on two different physical airway models.
The results are a useful lesson in reading simulation research.
The test
Single-centre, single-blinded randomised controlled trial. Each trainee had up to two hours to practise full, orderly and timely segmental navigation using pre-defined mastery learning protocols.
Full meant diagnostic completeness across 18 segments. Orderly meant structured progress through them. Timely meant procedure time. Mean intersegmental time was also recorded.
Experts assessed recorded complete test bronchoscopies on two different endobronchial airway training models: a BronchoBoy and a Koken. Cognitive load during training was compared using NASA-TLX.
Using two test models is the design choice that makes this trial interesting, because the AR group had trained on one of them.
The numbers
| Test and measure | Result | p |
|---|---|---|
| BronchoBoy: diagnostic completeness | VR better by median 1 segment | 0.0022 |
| BronchoBoy: structured progress | median difference 1 | 0.6887 |
| BronchoBoy: procedure time | 46 s | 0.8119 |
| BronchoBoy: mean intersegmental time | 2.15 s | 0.9760 |
| Koken: procedure time | AR faster by 48 s | 0.0377 |
| Koken: completeness, progress, intersegmental time | no difference | |
| Training time | VR 90 +/- 30 min, AR 80 +/- 30 min | 0.0468 |
| Mental demand (NASA-TLX) | VR 15 points lower | 0.023 |
| Physical demand (NASA-TLX) | VR 35 points higher | 0.004 |
The result
On the first test model, the BronchoBoy, VR trainees reached marginally more airway segments – a median difference of one, at p = 0.0022. Everything else on that model was indistinguishable.
On the second model, the Koken, the AR group was 48 seconds faster. The AR group had trained on the Koken. Completeness, structured progress and intersegmental time showed no difference.
The VR group also trained for ten minutes longer on average, 90 against 80 minutes, which is a confound for the completeness result.
The workload findings were the cleanest in the study. VR training carried 15 fewer NASA-TLX points of mental demand and 35 more points of physical demand.
The authors conclude that training by novices on either simulator led to thorough, timely and moderately ordered test segmental navigation, with VR portending slightly more thorough navigation, lower mental but higher physical demand, and longer time spent training.
How this compares to what came before
Simulation training for bronchoscopy is well established and the evidence that it accelerates competency is solid. What was missing was any head-to-head comparison of the two modalities.
Most comparisons in surgical simulation pit a simulator against no simulator, or against traditional apprenticeship. Those trials show large effects, because practice beats no practice. Head-to-head modality comparisons are rarer and almost always produce the kind of small, mixed result reported here.
That pattern is worth internalising. Once both options involve deliberate practice with feedback, the difference between delivery technologies tends to be small, and a trial of thirty people will struggle to find it.
What it means
The transfer finding is the most practically useful thing in the paper, and it is not the one in the abstract’s conclusion. The AR group was faster on the model they trained on. That is practice specificity – you get good at the thing you practised on – and it is a known phenomenon in motor learning.
For training programmes, this means that a simulator’s apparent superiority may partly reflect how closely the test resembles the training. A programme choosing a simulator should test trainees on something neither group has seen.
The workload finding has a different kind of use. Fifteen points of mental demand is substantial, and a trainee with more cognitive headroom can attend to more. Thirty-five points of physical demand in the other direction suggests VR was more tiring to operate, which matters for session length.
Neither simulator is clearly better at producing competent bronchoscopists. Both work. That is a reasonable answer to a question the field had not asked.
Business ideas from this paper
Transfer-neutral assessment models. If a simulator’s advantage depends on the test model, there is a market in a standard assessment mannequin that no vendor trains on. Training programmes and royal colleges would both buy it.
Workload instrumentation for simulators. NASA-TLX is a questionnaire. A simulator that captures workload proxies continuously – hesitation, correction rate, grip force – would give programmes a live view of trainee load.
Mastery-learning protocol licensing. The trial used pre-defined mastery learning protocols, which are the part that makes simulator training work. Packaging validated protocols separately from hardware is a software business with better margins than the mannequins.
Fatigue-aware session design. The 35-point physical demand gap suggests VR sessions should be shorter. Scheduling guidance grounded in measured workload is a small consulting product for training centres.
What could be wrong with it
Thirty trainees, fifteen per arm. Every comparison here rests on small numbers and a median difference of one airway segment is a fragile finding.
The VR group trained ten minutes longer, which undermines the one significant completeness result in VR’s favour.
The AR group’s speed advantage was on the model they had trained on, which the paper reports but does not fully discount.
The outcomes are simulator navigation scores. No patient was bronchoscoped, and transfer from simulator performance to clinical competence is assumed rather than measured.
Single centre, one training protocol, one pair of commercial simulators.
How sure can you be?
We score this 8 out of 10.
It is a peer-reviewed single-blinded randomised controlled trial in Respiration, with expert assessment of recorded test procedures on two independent physical models, and named institutions at Imperial College London, the Copenhagen Academy of Medical Education and Simulation at the University of Copenhagen, and Chelsea and Westminster Hospital NHS Foundation Trust.
It loses points because the data are not openly deposited beyond what is in the article and supplementary files, and because no independent site replicated the comparison.
The design is sound and the sample is small. We would report this as “no clear winner”, which is also what the authors report.
Do this today
If you run a simulation training programme, test your trainees on a model they did not train on. This trial shows how much of an apparent advantage can come from familiarity.
If you are buying simulators, ask the vendor for head-to-head data against the competing modality rather than against no training. Almost nobody has it, and that tells you something.
If you teach bronchoscopy, the mastery-learning protocol matters more than the hardware choice. Both arms here produced thorough, timely navigation, and the protocol was the same in both.
Source: Virtual Reality versus Augmented Reality Bronchoscopy Simulation-Based Training: A Randomised Controlled Trial to Compare Navigation Performance and Cognitive Load in Novice Bronchoscopists, Respiration, March 2026. PMID 41861064 · doi.org ().
Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.
- The authors state the premise had not been tested: whether AR offers advantages over VR in bronchoscopy training was unknown.
- VR beat AR marginally on diagnostic completeness on the first model, by a median of 1 segment.
- AR was 48 seconds faster on the second model - the one AR had trained on - which looks like practice specificity, not an AR advantage.
- VR carried lower mental demand and higher physical demand, which is a usability finding rather than a skills one.
Both VR and AR bronchoscopy simulators are commercially available and in use for training; this trial compares them directly under a mastery-learning protocol.
Questions people ask
What is the difference between VR and AR here?
VR training is fully virtual - the trainee navigates a simulated airway on a screen or headset. AR overlays the airway anatomy onto a physical model the trainee is actually manipulating.
Who won?
Nobody clearly. VR was marginally better on one measure on one model, AR was faster on the model it had trained on, and most comparisons showed no difference.
What is NASA-TLX?
A standard workload questionnaire measuring mental demand, physical demand, temporal demand, performance, effort and frustration. It is the usual instrument for comparing how taxing two interfaces are to use.
Why only thirty trainees?
Bronchoscopy training cohorts are small, and the trial used a mastery-learning protocol with expert-assessed test bronchoscopies on two physical models, which is labour-intensive. Thirty is typical for this kind of study and still too few for confident comparisons.
Mark Alex
Mark Alex is the founder and Managing Director of Real Biz Digital, a technology company operating out of Nairobi since 2018. He works in agentic AI and the Model Context Protocol, AI governance, enterprise software architecture and cybersecurity. He holds an MSc in Mechatronical Engineering from Obuda University in Budapest and a BSc in IT, Forensic Technology and Cybercrime, from USIU-Africa in Nairobi, and has published IEEE conference research on an AI-powered digital twin for greenhouse systems. He is the author of seven books. Between 2020 and 2024 he mentored more than 200 university students and interns in Nairobi. He writes every Just Out Tech article from the original research paper.